VisionWorks/OpenVX user kernels in immediate mode?

Robotics & Edge Computing Jetson & Embedded Systems Jetson TX2
1

The VisionWorks documentation, at top of ./NVIDIA_VisionWorks_Reference/nvvwx_docs
/group__group__user__kernels.html says this:

This seems to imply that a user created kernel can be run in immediate mode, but I can not find a single example of how to make this happen…exactly what is the aforementioned “interface”?

Specifically, I’m trying to get a small piece of code which is run only once to be GPU accelerated. (I’m currently building up a large remap within a double nested for-loop. It takes several seconds to run. Once its built, the actual vxRemapNode runs just fine in the downstream processing)

2

Hi,

You can find a custom kernel example in this topic:
[url]https://devtalk.nvidia.com/default/topic/1001961/jetson-tx1/visionworks-kernel-of-custom-node/[/url]

Thanks.

3

That example doesnt really answer my question at all. I have found lots of examples online in creating custom kernels to be used as vx_nodes in a graph. My question is specifically about executing a custom kernel in immediate mode, as the docs imply is possible.

Maybe for my particular desire this isn’t necessary and registering a kernel is just useless overhead.

4

Hi,

We don’t have an example for your requirement but it should work.
You can modify from the link shared in comment #2.

Graph Mode is immediate kernel plus some node handling.
It can give you some hint to implement it.

Thanks.